An
Analysis Of Dual Shuttle
Automated Storage/Retrieval Systems
Adhinarayan Keserla
Brett A. Peters
August 1, 1994
This working paper is not to be copied, quoted, or cited without the permission of the authors. Address correspondence to Brett A. Peters, Dept. of Industrial Engineering, Texas A&M University, College Station, TX 77843-3131 or email to bpeters@tamu.edu
An Analysis Of Dual Shuttle Automated Storage/Retrieval Systems
This paper addresses the throughput improvement possible with the use of a dual shuttle automated storage and retrieval system. With the use of such a system, travel between time in a dual command cycle is virtually eliminated resulting in a large throughput improvement. The dual shuttle system is then extended to perform an equivalent of two dual commands in one cycle in a quadruple command mode (QC). A heuristic that sequences retrievals to minimize travel time in QC mode is developed. Monte Carlo simulation results are provided for evaluating the heuristic's performance and show that it performs well, achieving large throughput improvements compared with that of the dual command cycle operating under the nearest neighbor retrieval sequencing heuristic.
Automated Storage/Retrieval Systems Design; Automated Storage/Retrieval Systems Operation; Material Handling Systems; Performance Modeling and Analysis
Automated storage/retrieval systems (AS/RS) are widely used in warehousing and manufacturing applications. A typical unit load AS/RS consists of storage racks, S/R machines, link conveyors, and input/output (I/O) stations. An important system performance measure is the throughput capacity of the system. The throughput capacity for a single aisle is the inverse of the mean transaction time, which is the expected amount of time required for the S/R machine to store and/or retrieve a unit load. The service time for a transaction includes both S/R machine travel time and pickup/deposit time. This time typically depends on the configuration of the storage rack and the S/R machine specifications.
The mean transaction time has been estimated by Bozer and White [1] for single command and dual command cycles for randomized storage and retrieval with different I/O configurations. Reducing the mean transaction time is critical for increasing the throughput of the AS/RS. One alternative is to reduce the dimensions of the rack which will decrease the mean transaction time. However, this reduction requires having more aisles to satisfy the same storage space requirements, hence the solution may not be cost effective.
Han et al. [2] improved the throughput capacity of the AS/RS through sequencing retrievals. Intelligently sequencing the retrievals can reduce unproductive travel between time when the S/R machine is traveling empty and thereby increase the throughput. They develop an expression for the maximum possible improvement in throughput if travel between is eliminated for an AS/RS that is throughput bound and operates in dual command mode. In essence, this means that if the S/R machine travels in a single command path but performs both a storage and a retrieval operation, the above throughput improvement could be obtained. With the existing S/R machine design, achieving this maximum throughput improvement is virtually impossible. Their analysis indicated, however, that for a typical AS/RS operating with 100% dual commands, the nearest neighbor heuristic could yield a 60% reduction in travel between times corresponding to a 12% increase in throughput. In large systems, a 12% increase in throughput could lead to the elimination of an aisle and the cost associated with the S/R machine.
In this paper, we analyze an alternative design of the S/R machine that has two shuttles instead of one as in a regular AS/RS. The new design eliminates the travel between the storage and retrieval points and performs both a storage and a retrieval at the point of retrieval, thereby achieving the maximum throughput increase calculated by Han et al. [4]. We extend this configuration to a dual shuttle AS/RS that has a higher throughput capacity than existing AS/R systems. We also provide heuristics for minimizing travel time by sequencing retrievals when the S/R machine operates in dual shuttle mode and give simulation results testing these heuristics. We then provide alternative methods for minimizing cycle times and point to some future directions for research.
In addition to the research work mentioned above, Elsayed and Unal [2] provide order batching algorithms for order retrieval in a multiple address AS/RS. Here an S/R machine cycle involves storing and retrieving multiple items in a single tour through several locations. They provide cycle time expressions for this multi-address tour time. The multi-address cycle is similar to a dual shuttle cycle when the number of visit locations is three. Their work, however, is intended for a different application, differs in storage and retrieval processing, and is therefore not directly applicable to the analysis of a dual shuttle system design.
The dual shuttle AS/RS is a new design aimed at improving S/R machine performance. Recently, several vendors have begun to offer such designs indicating that they may be a cost effective alternative in situations where high throughput is required. However, most studies on AS/R systems have been based on a single shuttle design. In our analysis of the dual shuttle AS/RS performance, we build upon these previous research results.
A typical unit-load AS/RS has an S/R machine operating in each aisle of the system. The S/R machine has a mast which is supported at the floor and the ceiling and travels horizontally within the aisle. Connected to this mast is a shuttle mechanism that carries the unit load and moves vertically up and down the mast. The shuttle mechanism also transfers loads in and out of storage locations in the rack. Figure 1 provides an illustration of the single shuttle S/R machine.
Figure 1. Single Shuttle S/R Machine Design
A typical single shuttle AS/RS can perform a single command cycle or a dual command cycle. A single command cycle consists of either a storage or a retrieval. For a storage, the time consists of the time to pickup the load at the I/O point, travel to the storage/retrieval point, deposit the load at that point, and return to the I/O point. The time for a retrieval is developed similarly.
A dual command cycle involves both a storage and a retrieval in the same cycle. The cycle time involves the time to pickup the load at the I/O point, travel to the storage location, place the load in the rack, travel empty to the retrieval location, retrieve a load, return to the I/O point, and deposit the load at the I/O point.
If we critically analyze the dual command cycle of the S/R machine (shown by the solid line in Figure 2), a potential open location for a future storage is created when a retrieval is performed. Furthermore, if both a retrieval and a storage are performed at the same point, the travel between time (TB) is eliminated, and the travel time will be equal to the single command travel time. With the existing AS/RS design, this mode of operation is not possible; therefore, an alternative to the S/R machine, a dual shuttle R/S machine, is proposed.
Figure 2. Dual Command Travel Paths of S/R and R/S Machines
Consider an S/R machine with two shuttle mechanisms instead of one. This new S/R machine could now carry two loads simultaneously. Each shuttle mechanism could operate independently of the other, so that individual loads can still be stored and retrieved. An illustration of the dual shuttle S/R machine is shown in Figure 3. This new S/R machine would operate as described below.
Figure 3. Dual Shuttle S/R Machine Design
The S/R machine picks up the item to be stored from the I/O point, loads it into the first shuttle, and moves to the retrieval location (as opposed to an open location in a regular AS/RS dual command operation). After reaching the retrieval location, the second shuttle is positioned to pickup the item to be retrieved. After retrieval, the S/R machine positions the first shuttle and deposits the load. The S/R machine then returns to the I/O point. The operation can easily be seen as a single command operation plus a small travel time for repositioning the S/R machine between the retrieval and storage (as well as the additional pickup and deposit time associated with the second load). Therefore, the S/R machine now operates as an R/S machine performing a retrieval first then a storage in a dual command cycle.
Since the R/S machine has two shuttles, the position of the shuttles has a role in the operation of the system. The two alternative shuttle configurations are the vertical shuttle design illustrated in Figure 3 and the horizontal shuttle design where the shuttles and hence the loads would sit side by side between the masts of the S/R machine. With two shuttles, the R/S machine is able to perform a dual command cycle at one location in the rack. This operation is accomplished by first retrieving the load onto the empty shuttle, transferring the second shuttle into position, and storing the load into the empty location in the rack. However, the choice of shuttle configuration does not impact the analysis in this paper.
To perform these operations, the R/S machine must move the second shuttle into position after the first shuttle has completed the retrieval. Due to the small distance involved, the R/S machine will use a slower creep speed for positioning, but this travel time is generally small. Furthermore, an amount of creep time is usually included in the pickup and deposit time to account for this required positioning. A second design characteristic is that additional clearance beyond the first and last row and column of the rack must be provided for overtravel of the R/S machine to accommodate both shuttle mechanisms.
To estimate the throughput improvement by the dual shuttle system over existing designs, we use the expressions for single command and dual command cycle times developed by Bozer and White [1] and the tabulated values for the nearest neighbor heuristic from Han et al. [4]. In developing the expressions, the authors in [1] and [4] made several assumptions. The same assumptions hold for the new design and include the following.
Bozer and White [1] have developed expected cycle time expressions for single and dual command cycles. They assumed randomized storage and FCFS processing of both storages and retrievals. The following equations were developed [1].
(1)
(2)
(3)
where
The travel time expression for the R/S machine
performing a dual command cycle will be E(SC) plus
the creep time, tc, required to reposition the second shuttle in front of
the rack opening. This creep time, tc, can be approximated by 
(for a horizontal shuttle configuration)
where w is the width of a rack opening, although it is
likely to be larger than this value. Then, the expected dual
command cycle time for the R/S machine, E(DC)r/s, is given by
E(DC)r/s = E(SC) + tc (4)
Let E(DC)s/r be the travel time for a dual command cycle from
Bozer and White [1] shown in equation (3) above, and let 
be the travel time for
a dual command cycle using the nearest neighbor retrieval
sequencing heuristic with a block size of n retrievals and
m open locations in the rack from Han et al. [4].
The cycle time, CT, will add the pickup and deposit time
to the travel time and will be denoted by the subscripts r/s,
s/r, and NN for the R/S machine, the S/R machine,
and the S/R machine using nearest neighbor retrieval sequencing,
respectively. Similarly, the throughput, µ, will be denoted by
the appropriate subscript.
Assuming b = 1, T = 1.0, tc = 0.0111 minutes, tp/d = 0.2 minutes
E(DC)r/s = 1.3333 + 0.0111 = 1.3444
CTr/s = 1.3444 + 4(0.2) = 2.1444 minutes
µr/s = 2(60 / 2.1444) = 55.95 operations per hour .
E(DC)s/r = 1.8 minutes
CTs/r = 1.8 + 4(0.2) = 2.6 minutes
µs/r = 2(60 / 2.6) = 46.15 operations per hour
Using the nearest neighbor heuristic [4] with m=10 and n=25
= 1.4372 minutes (From Table 3 in Han et al. [4])
CTNN = 1.4372 + 4(0.2) = 2.2372 minutes
µNN = 2(60 / 2.2372) = 53.63 operations per hour
Thus, the throughput improvement for the R/S machine over a FCFS S/R dual command operation is given by r/s, while the throughput improvement from sequencing retrievals using the nearest neighbor heuristic over the FCFS S/R dual command cycle time is NN.
The above calculations clearly show the throughput improvement for the R/S machine, r/s, is very close to the maximum possible improvement of 22% [4] and has a higher throughput than the dual command cycle with the nearest neighbor heuristic with 10 open storage locations and a block size of 25 retrievals. Along with the higher throughput, there are other advantages of using the R/S machine: theoretically a 100% utilization of the rack-face is possible (i.e., no open locations are needed as long as dual command cycles are always performed) and no retrieval sequencing is required (i.e., all retrievals will be processed on a FCFS basis without affecting the system performance).
The new design of the S/R machine has two shuttles and therefore could be operated as a dual shuttle system: carrying two loads and depositing them, retrieving two loads, and returning to the I/O point to deliver them as shown in Figure 4. The above operation can be performed by storing and retrieving the loads at four different locations. Therefore, the travel time would consist of the time for a single command travel plus three travel between times. To more efficiently perform the 4 operations, a retrieval and storage performed at one location is interspersed with a dual command operation. This mode of operation, termed the quadruple command (QC) cycle, eliminates one travel between and is more efficient than the previous mode mentioned above (see Figure 5). The QC cycle can be performed with storages at randomized locations and retrievals processed in a first-come-first-served (FCFS) manner. However, by intelligently sequencing the retrieval list, the travel time in performing the four operations can be significantly reduced. This type of analysis was used by Han et al. [4] to improve the throughput of a single load AS/RS. In our paper, we build on the results of their analysis. The notation and the assumptions mentioned in section 2.2. still hold, except that multiple stops of the S/R machine are now allowed.
Figure 4. S/R Machine Path Performing Four Operations At Four Locations.
Figure 5. S/R Machine Path Performing Four Operations At Three Locations.
The QC mode of operation requires the S/R machine to visit three locations on the storage rack. An efficient way to perform this command is to sequence the retrieval list such that the points in the triple (S1, R1, R2) are close together. The problem now is how to optimally find this triple, that is, the set of three locations out of the available storage and retrieval locations that are closest together.
The problem of optimally sequencing a given list of retrievals with a single open location for initial storage operations using a single shuttle AS/RS is equivalent to a traveling salesman problem (TSP). Since there is only a single open location, the storage point is always known and therefore the only decision is what order to visit the retrieval points. The travel distance (or time) between any pair of retrievals is known since the previous retrieval point will be the storage point for the next cycle. (See [4] for additional information.) The traveling salesman problem belongs to the class of NP complete problems [3], which is strong evidence that no efficient optimal solution algorithm can be developed (although several efficient heuristic procedures have been developed for the TSP).
In this paper, a dual shuttle system with multiple storage locations is considered. This situation increases the complexity of the problem beyond a simple TSP. In the absence of an optimal algorithm, we consider heuristics that seek to minimize travel times. These heuristics exploit the structure of the problem to obtain good solutions in a computationally efficient manner. Before discussing the heuristic, we discuss the desirable characteristics of a heuristic that should lead to good solutions.
In the next section, we discuss the minimum perimeter heuristic, which satisfies these characteristics, and develop upper and lower bounds on the expected quadruple command travel time.
Figure 6. S/R Machine Path With And Without Cross-Over
To sequence the retrievals, we would like to find three points
that are close together. If we consider the problem of finding
three such points, one possibility is to find a triangle with the
minimum perimeter, where the vertices of the triangle are the
visit locations of the S/R machine and the edges represent the
travel between distances. The next question is how efficiently
can we find the smallest perimeter triangle. The heuristic needs
to be implemented in real time; therefore, it should be a
computationally efficient procedure. We consider two methods. One
simple method enumerates all possible triangles between the
points and returns the smallest perimeter triangle. The number of
triangles computed depends upon m and n and is
equal to 
for
every cycle of the S/R machine, where m is the number of
initial open storage locations in the rack and n is the
number of retrievals in the block to be sequenced. The other
method is the minimum perimeter heuristic (MP). It is an
efficient method that resembles the nearest neighbor heuristic
discussed in [4] but yet is distinctly different.
The MP heuristic procedure is simple to operate for sequencing retrievals. Let R be the set of n retrievals and S be the set of initial open locations. The MP heuristic operates as follows.
While R {}
The steps in the above procedure can be described as follows, step 1 is a direct implementation of the nearest neighbor technique. Step 2 is a modification of the nearest neighbor technique, here we choose r2 which is closer to both r1 and s1 rather than just r1. Step 3 ensures that there is no cross over of the S/R path during its cycle. Step 4 chooses the second storage to perform in the QC cycle. This choice doesn't impact the current cycle's time, but may effect the cycle time of future cycles. It can be implemented in a number of methods, including random choice, FCFS (or next in queue), or largest sum of travel between with r R - {r1, r2}. The final step performs the QC operation and updates the retrieval list and set of open locations.
The decrease in travel time due to retrieval sequencing will depend on the retrieval block size and the number of open locations. It is obvious that as the block size and number of locations increases the travel time decreases. There will be a limit, however, to the number of open locations that can be provided on the rack face due to storage requirement constraints and a limit on the retrieval block size due to the maximum wait time that can be imposed on a retrieval before it can be processed.
In our simulations, we use values of m ranging from 1 to 10 and retrieval block size from 2 to 24 in steps of 2 due to the nature of the QC command. Retrieval lists and open locations were generated by Monte Carlo sampling.
In the Monte Carlo sampling procedure, points were randomly generated in a continuous unit square, which implies a shape factor of b = 1. The shape factor b was chosen to be one since this value of b is optimal [1], i.e., it minimizes travel time; however, the results of this analysis are equally applicable for other shape factor values.
For each sample, n random retrieval points were generated for each cycle whereas m random open locations were generated only at the beginning of the first cycle, i.e., dynamic simulation of the system was performed [4]. The open locations at the end of each cycle were used for subsequent cycles. The results of the simulation are shown in Table 1. The table shows that the travel times are not symmetric with respect to m and n. The travel time is more sensitive to n than to m. This result is opposite to that obtained in the Han et al. paper [4] where the travel times are more sensitive to m than to n. The contradiction between the two results can be explained by noting that in the QC cycle two retrievals and two storages are performed but the S/R machine visits two retrieval locations and only one storage location, storing the other load in an opening created by one of the retrievals from that cycle. Therefore, the retrieval block size has a larger impact on the magnitude of the cycle time than does the number of open locations.
Two measures of performance were considered in comparing the minimum perimeter heuristic method with the enumeration method.
We simulated the two methods for 1000 cycles for two combinations of m = 5 (# of open locations) and n = 10 (retrieval block size) on an 80486 class machine. The travel times obtained from both the methods were approximately the same with minor differences in only the 3rd decimal position. However, the enumeration method took 3 minutes of CPU time while the minimum perimeter heuristic took 0.38 minutes. These results show that both methods could be implemented in real time but the minimum perimeter heuristic method is probably preferred.
Table 1. Results of Monte Carlo Simulation for Quadruple Command Cycle
To reduce the travel time further, an attempt was made to take advantage of the no-cost zone [4] (see Figure 7). The no-cost zone is the area in the rack that the S/R machine can cover without increasing the travel time from the point (R1, S2) to the I/O point. Because of simultaneous horizontal and vertical travel of the S/R machine, it can visit any location in the shaded area without increasing the travel time in returning to the I/O point (ignoring acceleration and deceleration effects).
Figure 7. Illustration of No-Cost Zone
For the dual shuttle system, the no-cost zone was exploited as follows: once a storage and retrieval pair were selected using the nearest-neighbor heuristic, the 2nd retrieval was chosen from the no-cost zone using the 1st retrieval point as a reference. Note that, since the 2nd retrieval is performed in the no-cost zone, open locations will move closer to the I/O point exhibiting good dynamic performance. In the Han et al. paper [4], the opposite was true for a single shuttle S/R machine operating under the no-cost zone heuristic; that is, the open locations moved further away from the I/O point increasing the cycle time over time and resulting in poor dynamic performance.
For the dual shuttle system, simulation was performed for two cases (m=10, n=20 and m=5, n=10) for 1000 cycles. The simulation results showed that the open locations tend to be closer to the I/O point. Surprisingly, the travel time increased slightly. This result is due to not choosing the pair of retrievals to minimize distance, which further illustrates the increased sensitivity of the cycle time to n as opposed to m, the number of open locations. For m=10 and n=20 the cycle time was equal to 1.574 and for m=5 and n=10 the cycle time was 1.666. These values can be compared to the corresponding values in Table 1. Although the no-cost zone exhibits good dynamic behavior, it doesn't appreciably improve overall system performance. Therefore, further analysis was not pursued.
We could consider other possible heuristics (e.g., heuristics based on space filling curves) which may yield lower travel times. However before considering other heuristics, it is prudent to determine how far above optimal the minimum perimeter heuristic is, since if it is sufficiently close then there may be no need to consider other heuristics. Not having a closed form expression for the QC travel time, upper and lower bound expressions for the expected cycle time were developed. The analysis extends the results found in the Han et al. paper [4].
The upper bound, E(QC)UB, is computed by simply applying the nearest neighbor
heuristic twice. That is, find a storage and retrieval pair that
minimizes travel between. Then, choose a second retrieval that
minimizes travel between with the first retrieval. As before, 
is the dual command
cycle travel time for the single shuttle S/R machine operating
under the nearest neighbor retrieval sequencing heuristic and 
is the travel between
time for the nearest neighbor retrieval sequencing heuristic with
n-1 retrievals to choose from and a single open location.
(5)
The lower bound, E(QC)LB, is computed by recognizing that the minimum travel time occurs when m and n are maximum. This result occurs because as retrievals are processed the number of selections from R gets reduced, consequently the travel between time increases. The minimum travel between time occurs initially when m and n are maximum and the second travel between takes place in the no-cost zone. Therefore, the appropriate expression for the lower bound is
(6)
where the E(Zk) value is the expected value of the smallest of k random distances [4], and pncz is the probability that the second retrieval is not in the no-cost zone.
The probability that a retrieval is in the no-cost zone can be estimated as follows. In [4], Han et al. state that the expected value of the area of the no-cost zone is 0.125 for b=1. It follows then that the probability of zero retrievals in the no-cost zone is given by e-0.125(n-1) using the result that the number of locations in a given size area will follow a Poisson distribution [4].
The expected values of the components in the right-hand side of equations (5) and (6) are taken from Tables 1-5 of Han et al. [4]. Two pairs of m and n were chosen for illustrating the bounds. The results are summarized in Table 2. In terms of throughput, the minimum perimeter heuristic is within 5.0% and 8.3% of the maximum possible throughput. The results from Table 2 show that the cycle times are not far from the lower bound and the resulting throughput values are also close to the maximum possible, verifying the performance of the minimum perimeter heuristic. Theorectically, better heuristics are possible, but the improvement over the minimum perimeter heuristic would be small, less than 10% in most cases.
Table 2. Comparison of Minimum
Perimeter Heuristic Performance With Upper and Lower Bounds
Optimal values of m and n that give the best possible travel times considering the number of open locations and the size of the retrieval list are considered. Optimal values of m and n should be determined not only with respect to travel time but also other factors such as cost per cubic feet of storage space, maximum wait time limit on the retrievals, etc. Figure 8 shows a graph of the cycle time for various values of n, the retrieval block size, with three different values of m, the number of open locations. The graph shows that the decrease in travel time changes rapidly until n = 16, after that point the decrease is minimal. The graph also shows that increasing the number of open locations does not decrease the travel time as much as increasing the retrieval block size. Therefore, if storage space is a premium, then the retrieval block size can be increased to get the same reduction in travel time. For example, from the graph the travel time of m = 5 and n = 24 is equivalent to the travel time with m = 10 and n = 12.
Based on these experimental results, a range of values for m and n from 5 to 7 and 14 to 20, respectively, would appear to be the best choices for achieving good throughput performance while maintaining high utilization of the rack and not drastically increasing the waiting time for retrievals. However, these values should be adjusted for the specifics of each particular situation.
Comparison of the quadruple command cycle with a dual command cycle using nearest neighbor retrieval sequencing shows large improvements in throughput. For example, consider the following case with m = 10, n = 20, and tp/d = 0.2 minutes.
= 1.5048 + 8(0.2) = 3.1048
= 1.4429 + 4(0.2) = 2.2429
= 77.2997 operations per hour
= 53.5022 operations per hour
Figure 8. Quadruple Command Cycle Time Versus Size of n and m
A throughput improvement of 44% is dramatic. This improvement amounts to eliminating one out of every three aisles. Large AS/RS may contain several aisles, and hence, large savings could result from the use of dual shuttle systems. Even though a dual shuttle S/R machine costs more that a single shuttle S/R machine, this increase may be less than the increase in savings due to the elimination of aisles. An economic analysis is needed to confirm this possibility for each situation.
It may also not be necessary that all aisles have dual shuttle S/R machines. Aisles having high turnover items assigned to them could have dual shuttle S/R machines; those aisles with lower throughput could use the single shuttle S/R machine. Clearly, the elimination of aisles depends on the required storage requirements. It may be possible, however, with the dual shuttle system to have fewer, larger aisles and not suffer a degradation in throughput performance. All of these factors need to be considered to determine if dual shuttle systems are cost effective alternatives. From the throughput standpoint, they look promising and deserve further consideration. The analysis in this paper can be used to evaluate the alternative scenarios for any particular situation to determine the proper system design.
This paper performs an analysis of dual shuttle automated storage and retrieval systems. Several contributions have been made including the following.
The dual shuttle system shows promise for situations requiring high throughput. The main disadvantage with the new design is the extra cost of the S/R machine. An economic evaluation is needed to determine if it is appropriate for a particular situation. However, based on throughput performance, the dual shuttle design appears promising.
The concept of dual shuttle systems can also be extended to other material handling systems, such as rotary rack carousels. Furthermore, research is needed to consider other storage strategies, such as class based storage policies, to examine their impact on throughput in conjunction with the dual shuttle design. This paper provides a framework for analyzing dual shuttle AS/RS, and it provides a foundation for other material handling research related to this concept.
[1] Bozer, Y.A. and J.A. White, "Travel-Time Models for Automated Storage/Retrieval Systems," IIE Transactions, Vol. 16 , No. 4, 1984, 329-338.
[2] Elsayed, E.A. and O.I. Unal, "Order Batching Algorithms and Travel Time Estimation for Automated Storage/Retrieval Systems," International Journal of Production Research, Vol. 27, No. 7, 1989, 1097-1114.
[3] Garey, M. and D.S. Johnson, Computers and Intractability: A Guide to the Theory of Computing, Freeman Press, San Francisco, CA, 1979.
[4] Han, M.H., L.F. McGinnis, J.S. Shieh, and J.A. White, "On Sequencing Retrievals In An Automated Storage/Retrieval System," IIE Transactions, March 1987, 56-66.
Adhinarayan A. Keserla is a Quality Engineer at Blom Industries, Inc. in Mt. Clemens, MI. He has a M.S. in Industrial Engineering from Texas A&M University and a B.S. in Industrial and Production Engineering from the Bangalore University. His interests include robust design of manufacturing systems, material handling, application of statistical techniques for process improvement, and ISO 9000 quality systems. Mr. Keserla is a member of IIE and ASQC.
Brett A. Peters is an Assistant Professor in the Industrial Engineering Department at Texas A&M University. He has a Ph.D. and M.S. in Industrial Engineering from the Georgia Institute of Technology and a B.S.I.E. from the University of Arkansas. His research areas include facilities design, material handling, and design and analysis of manufacturing systems. Dr. Peters is a member of SME, IIE, and ORSA.